diff --git a/ormar/__init__.py b/ormar/__init__.py
index b2c7020..328e894 100644
--- a/ormar/__init__.py
+++ b/ormar/__init__.py
@@ -56,7 +56,7 @@ from ormar.fields import (
 )  # noqa: I100
 from ormar.models import Model
 from ormar.models.metaclass import ModelMeta
-from ormar.queryset import QuerySet
+from ormar.queryset import OrderAction, QuerySet
 from ormar.relations import RelationType
 from ormar.signals import Signal
 
@@ -106,4 +106,5 @@ __all__ = [
     "BaseField",
     "ManyToManyField",
     "ForeignKeyField",
+    "OrderAction",
 ]
diff --git a/ormar/models/mixins/prefetch_mixin.py b/ormar/models/mixins/prefetch_mixin.py
index d8ee350..440052a 100644
--- a/ormar/models/mixins/prefetch_mixin.py
+++ b/ormar/models/mixins/prefetch_mixin.py
@@ -2,7 +2,7 @@ from typing import Callable, Dict, List, TYPE_CHECKING, Tuple, Type, cast
 
 from ormar.models.mixins.relation_mixin import RelationMixin
 
-if TYPE_CHECKING:
+if TYPE_CHECKING:  # pragma: no cover
     from ormar.fields import ForeignKeyField, ManyToManyField
 
 
@@ -18,10 +18,10 @@ class PrefetchQueryMixin(RelationMixin):
 
     @staticmethod
     def get_clause_target_and_filter_column_name(
-        parent_model: Type["Model"],
-        target_model: Type["Model"],
-        reverse: bool,
-        related: str,
+            parent_model: Type["Model"],
+            target_model: Type["Model"],
+            reverse: bool,
+            related: str,
     ) -> Tuple[Type["Model"], str]:
         """
         Returns Model on which query clause should be performed and name of the column.
@@ -51,7 +51,7 @@ class PrefetchQueryMixin(RelationMixin):
 
     @staticmethod
     def get_column_name_for_id_extraction(
-        parent_model: Type["Model"], reverse: bool, related: str, use_raw: bool,
+            parent_model: Type["Model"], reverse: bool, related: str, use_raw: bool,
     ) -> str:
         """
         Returns name of the column that should be used to extract ids from model.
diff --git a/ormar/models/model_row.py b/ormar/models/model_row.py
index f0a4a31..476e274 100644
--- a/ormar/models/model_row.py
+++ b/ormar/models/model_row.py
@@ -17,7 +17,7 @@ from ormar.models import NewBaseModel  # noqa: I202
 from ormar.models.helpers.models import group_related_list
 
 
-if TYPE_CHECKING:
+if TYPE_CHECKING:  # pragma: no cover
     from ormar.fields import ForeignKeyField
     from ormar.models import T
 else:
diff --git a/ormar/queryset/__init__.py b/ormar/queryset/__init__.py
index ebfab7b..11b072e 100644
--- a/ormar/queryset/__init__.py
+++ b/ormar/queryset/__init__.py
@@ -1,10 +1,20 @@
 """
 Contains QuerySet and different Query classes to allow for constructing of sql queries.
 """
+from ormar.queryset.actions import FilterAction, OrderAction
 from ormar.queryset.filter_query import FilterQuery
 from ormar.queryset.limit_query import LimitQuery
 from ormar.queryset.offset_query import OffsetQuery
 from ormar.queryset.order_query import OrderQuery
 from ormar.queryset.queryset import QuerySet, T
 
-__all__ = ["T", "QuerySet", "FilterQuery", "LimitQuery", "OffsetQuery", "OrderQuery"]
+__all__ = [
+    "T",
+    "QuerySet",
+    "FilterQuery",
+    "LimitQuery",
+    "OffsetQuery",
+    "OrderQuery",
+    "FilterAction",
+    "OrderAction",
+]
diff --git a/ormar/queryset/actions/__init__.py b/ormar/queryset/actions/__init__.py
new file mode 100644
index 0000000..088d68a
--- /dev/null
+++ b/ormar/queryset/actions/__init__.py
@@ -0,0 +1,4 @@
+from ormar.queryset.actions.filter_action import FilterAction
+from ormar.queryset.actions.order_action import OrderAction
+
+__all__ = ["FilterAction", "OrderAction"]
diff --git a/ormar/queryset/filter_action.py b/ormar/queryset/actions/filter_action.py
similarity index 71%
rename from ormar/queryset/filter_action.py
rename to ormar/queryset/actions/filter_action.py
index d2d8e45..43c71df 100644
--- a/ormar/queryset/filter_action.py
+++ b/ormar/queryset/actions/filter_action.py
@@ -1,11 +1,11 @@
-from typing import Any, Dict, List, TYPE_CHECKING, Type
+from typing import Any, Dict, TYPE_CHECKING, Type
 
 import sqlalchemy
 from sqlalchemy import text
 
 import ormar  # noqa: I100, I202
 from ormar.exceptions import QueryDefinitionError
-from ormar.queryset.utils import get_relationship_alias_model_and_str
+from ormar.queryset.actions.query_action import QueryAction
 
 if TYPE_CHECKING:  # pragma: nocover
     from ormar import Model
@@ -28,7 +28,7 @@ FILTER_OPERATORS = {
 ESCAPE_CHARACTERS = ["%", "_"]
 
 
-class FilterAction:
+class FilterAction(QueryAction):
     """
     Filter Actions is populated by queryset when filter() is called.
 
@@ -39,7 +39,18 @@ class FilterAction:
     """
 
     def __init__(self, filter_str: str, value: Any, model_cls: Type["Model"]) -> None:
-        parts = filter_str.split("__")
+        super().__init__(query_str=filter_str, model_cls=model_cls)
+        self.filter_value = value
+        self._escape_characters_in_clause()
+
+    def has_escaped_characters(self) -> bool:
+        """Check if value is a string that contains characters to escape"""
+        return isinstance(self.filter_value, str) and any(
+            c for c in ESCAPE_CHARACTERS if c in self.filter_value
+        )
+
+    def _split_value_into_parts(self, query_str: str) -> None:
+        parts = query_str.split("__")
         if parts[-1] in FILTER_OPERATORS:
             self.operator = parts[-1]
             self.field_name = parts[-2]
@@ -49,61 +60,6 @@ class FilterAction:
             self.field_name = parts[-1]
             self.related_parts = parts[:-1]
 
-        self.filter_value = value
-        self.table_prefix = ""
-        self.source_model = model_cls
-        self.target_model = model_cls
-        self.is_through = False
-        self._determine_filter_target_table()
-        self._escape_characters_in_clause()
-
-    @property
-    def table(self) -> sqlalchemy.Table:
-        """Shortcut to sqlalchemy Table of filtered target model"""
-        return self.target_model.Meta.table
-
-    @property
-    def column(self) -> sqlalchemy.Column:
-        """Shortcut to sqlalchemy column of filtered target model"""
-        aliased_name = self.target_model.get_column_alias(self.field_name)
-        return self.target_model.Meta.table.columns[aliased_name]
-
-    def has_escaped_characters(self) -> bool:
-        """Check if value is a string that contains characters to escape"""
-        return isinstance(self.filter_value, str) and any(
-            c for c in ESCAPE_CHARACTERS if c in self.filter_value
-        )
-
-    def update_select_related(self, select_related: List[str]) -> List[str]:
-        """
-        Updates list of select related with related part included in the filter key.
-        That way If you want to just filter by relation you do not have to provide
-        select_related separately.
-
-        :param select_related: list of relation join strings
-        :type select_related: List[str]
-        :return: list of relation joins with implied joins from filter added
-        :rtype: List[str]
-        """
-        select_related = select_related[:]
-        if self.related_str and not any(
-            rel.startswith(self.related_str) for rel in select_related
-        ):
-            select_related.append(self.related_str)
-        return select_related
-
-    def _determine_filter_target_table(self) -> None:
-        """
-        Walks the relation to retrieve the actual model on which the clause should be
-        constructed, extracts alias based on last relation leading to target model.
-        """
-        (
-            self.table_prefix,
-            self.target_model,
-            self.related_str,
-            self.is_through,
-        ) = get_relationship_alias_model_and_str(self.source_model, self.related_parts)
-
     def _escape_characters_in_clause(self) -> None:
         """
         Escapes the special characters ["%", "_"] if needed.
@@ -151,7 +107,7 @@ class FilterAction:
         sufix = "%" if "end" not in self.operator else ""
         self.filter_value = f"{prefix}{self.filter_value}{sufix}"
 
-    def get_text_clause(self,) -> sqlalchemy.sql.expression.TextClause:
+    def get_text_clause(self) -> sqlalchemy.sql.expression.TextClause:
         """
         Escapes characters if it's required.
         Substitutes values of the models if value is a ormar Model with its pk value.
diff --git a/ormar/queryset/actions/order_action.py b/ormar/queryset/actions/order_action.py
new file mode 100644
index 0000000..2173e24
--- /dev/null
+++ b/ormar/queryset/actions/order_action.py
@@ -0,0 +1,68 @@
+from typing import TYPE_CHECKING, Type
+
+import sqlalchemy
+from sqlalchemy import text
+
+from ormar.queryset.actions.query_action import QueryAction  # noqa: I100, I202
+
+if TYPE_CHECKING:  # pragma: nocover
+    from ormar import Model
+
+
+class OrderAction(QueryAction):
+    """
+    Order Actions is populated by queryset when order_by() is called.
+
+    All required params are extracted but kept raw until actual filter clause value
+    is required -> then the action is converted into text() clause.
+
+    Extracted in order to easily change table prefixes on complex relations.
+    """
+
+    def __init__(
+        self, order_str: str, model_cls: Type["Model"], alias: str = None
+    ) -> None:
+        self.direction: str = ""
+        super().__init__(query_str=order_str, model_cls=model_cls)
+        self.is_source_model_order = False
+        if alias:
+            self.table_prefix = alias
+        if self.source_model == self.target_model and "__" not in self.related_str:
+            self.is_source_model_order = True
+
+    @property
+    def field_alias(self) -> str:
+        return self.target_model.get_column_alias(self.field_name)
+
+    def get_text_clause(self) -> sqlalchemy.sql.expression.TextClause:
+        """
+        Escapes characters if it's required.
+        Substitutes values of the models if value is a ormar Model with its pk value.
+        Compiles the clause.
+
+        :return: complied and escaped clause
+        :rtype: sqlalchemy.sql.elements.TextClause
+        """
+        prefix = f"{self.table_prefix}_" if self.table_prefix else ""
+        return text(f"{prefix}{self.table}" f".{self.field_alias} {self.direction}")
+
+    def _split_value_into_parts(self, order_str: str) -> None:
+        if order_str.startswith("-"):
+            self.direction = "desc"
+            order_str = order_str[1:]
+        parts = order_str.split("__")
+        self.field_name = parts[-1]
+        self.related_parts = parts[:-1]
+
+    def check_if_filter_apply(self, target_model: Type["Model"], alias: str) -> bool:
+        """
+        Checks filter conditions to find if they apply to current join.
+
+        :param target_model: model which is now processed
+        :type target_model: Type["Model"]
+        :param alias: prefix of the relation
+        :type alias: str
+        :return: result of the check
+        :rtype: bool
+        """
+        return target_model == self.target_model and alias == self.table_prefix
diff --git a/ormar/queryset/actions/query_action.py b/ormar/queryset/actions/query_action.py
new file mode 100644
index 0000000..2c6ee84
--- /dev/null
+++ b/ormar/queryset/actions/query_action.py
@@ -0,0 +1,93 @@
+import abc
+from typing import Any, List, TYPE_CHECKING, Type
+
+import sqlalchemy
+
+from ormar.queryset.utils import get_relationship_alias_model_and_str  # noqa: I202
+
+if TYPE_CHECKING:  # pragma: nocover
+    from ormar import Model
+
+
+class QueryAction(abc.ABC):
+    """
+    Base QueryAction class with common params for Filter and Order actions.
+    """
+
+    def __init__(self, query_str: str, model_cls: Type["Model"]) -> None:
+        self.query_str = query_str
+        self.field_name: str = ""
+        self.related_parts: List[str] = []
+        self.related_str: str = ""
+
+        self.table_prefix = ""
+        self.source_model = model_cls
+        self.target_model = model_cls
+        self.is_through = False
+
+        self._split_value_into_parts(query_str)
+        self._determine_filter_target_table()
+
+    def __eq__(self, other: object) -> bool:  # pragma: no cover
+        if not isinstance(other, QueryAction):
+            return False
+        return self.query_str == other.query_str
+
+    def __hash__(self) -> Any:
+        return hash((self.table_prefix, self.query_str))
+
+    @abc.abstractmethod
+    def _split_value_into_parts(self, query_str: str) -> None:  # pragma: no cover
+        """
+        Splits string into related parts and field_name
+        :param query_str: query action string to split (i..e filter or order by)
+        :type query_str: str
+        """
+        pass
+
+    @abc.abstractmethod
+    def get_text_clause(
+        self,
+    ) -> sqlalchemy.sql.expression.TextClause:  # pragma: no cover
+        pass
+
+    @property
+    def table(self) -> sqlalchemy.Table:
+        """Shortcut to sqlalchemy Table of filtered target model"""
+        return self.target_model.Meta.table
+
+    @property
+    def column(self) -> sqlalchemy.Column:
+        """Shortcut to sqlalchemy column of filtered target model"""
+        aliased_name = self.target_model.get_column_alias(self.field_name)
+        return self.target_model.Meta.table.columns[aliased_name]
+
+    def update_select_related(self, select_related: List[str]) -> List[str]:
+        """
+        Updates list of select related with related part included in the filter key.
+        That way If you want to just filter by relation you do not have to provide
+        select_related separately.
+
+        :param select_related: list of relation join strings
+        :type select_related: List[str]
+        :return: list of relation joins with implied joins from filter added
+        :rtype: List[str]
+        """
+        select_related = select_related[:]
+        if self.related_str and not any(
+            rel.startswith(self.related_str) for rel in select_related
+        ):
+            select_related.append(self.related_str)
+        return select_related
+
+    def _determine_filter_target_table(self) -> None:
+        """
+        Walks the relation to retrieve the actual model on which the clause should be
+        constructed, extracts alias based on last relation leading to target model.
+        """
+        (
+            self.table_prefix,
+            self.target_model,
+            self.related_str,
+            self.is_through,
+        ) = get_relationship_alias_model_and_str(self.source_model, self.related_parts)
diff --git a/ormar/queryset/clause.py b/ormar/queryset/clause.py
index b5a3c5b..b98616d 100644
--- a/ormar/queryset/clause.py
+++ b/ormar/queryset/clause.py
@@ -3,7 +3,7 @@ from dataclasses import dataclass
 from typing import Any, List, TYPE_CHECKING, Tuple, Type
 
 import ormar  # noqa I100
-from ormar.queryset.filter_action import FilterAction
+from ormar.queryset.actions.filter_action import FilterAction
 from ormar.queryset.utils import get_relationship_alias_model_and_str
 
 if TYPE_CHECKING:  # pragma no cover
diff --git a/ormar/queryset/filter_query.py b/ormar/queryset/filter_query.py
index 4100f16..cb9b880 100644
--- a/ormar/queryset/filter_query.py
+++ b/ormar/queryset/filter_query.py
@@ -1,7 +1,7 @@
 from typing import List
 
 import sqlalchemy
-from ormar.queryset.filter_action import FilterAction
+from ormar.queryset.actions.filter_action import FilterAction
 
 
 class FilterQuery:
diff --git a/ormar/queryset/join.py b/ormar/queryset/join.py
index 4626fbd..f18c81d 100644
--- a/ormar/queryset/join.py
+++ b/ormar/queryset/join.py
@@ -14,11 +14,13 @@ from typing import (
 import sqlalchemy
 from sqlalchemy import text
 
-from ormar.exceptions import RelationshipInstanceError  # noqa I100
+import ormar  # noqa I100
+from ormar.exceptions import RelationshipInstanceError
 from ormar.relations import AliasManager
 
 if TYPE_CHECKING:  # pragma no cover
     from ormar import Model
+    from ormar.queryset import OrderAction
 
 
 class SqlJoin:
@@ -29,7 +31,7 @@ class SqlJoin:
         columns: List[sqlalchemy.Column],
         fields: Optional[Union[Set, Dict]],
         exclude_fields: Optional[Union[Set, Dict]],
-        order_columns: Optional[List],
+        order_columns: Optional[List["OrderAction"]],
         sorted_orders: OrderedDict,
         main_model: Type["Model"],
         relation_name: str,
@@ -89,7 +91,18 @@ class SqlJoin:
         """
         return self.main_model.Meta.alias_manager
 
-    def on_clause(self, previous_alias: str, from_clause: str, to_clause: str,) -> text:
+    @property
+    def to_table(self) -> str:
+        """
+        Shortcut to table name of the next model
+        :return: name of the target table
+        :rtype: str
+        """
+        return self.next_model.Meta.table.name
+
+    def _on_clause(
+        self, previous_alias: str, from_clause: str, to_clause: str,
+    ) -> text:
         """
         Receives aliases and names of both ends of the join and combines them
         into one text clause used in joins.
@@ -118,7 +131,7 @@ class SqlJoin:
         :rtype: Tuple[List[str], Join, List[TextClause], collections.OrderedDict]
         """
         if self.target_field.is_multi:
-            self.process_m2m_through_table()
+            self._process_m2m_through_table()
 
         self.next_model = self.target_field.to
         self._forward_join()
@@ -207,7 +220,7 @@ class SqlJoin:
             self.sorted_orders,
         ) = sql_join.build_join()
 
-    def process_m2m_through_table(self) -> None:
+    def _process_m2m_through_table(self) -> None:
         """
         Process Through table of the ManyToMany relation so that source table is
         linked to the through table (one additional join)
@@ -222,8 +235,7 @@ class SqlJoin:
 
         To point to through model
         """
-        new_part = self.process_m2m_related_name_change()
-        self._replace_many_to_many_order_by_columns(self.relation_name, new_part)
+        new_part = self._process_m2m_related_name_change()
 
         self.next_model = self.target_field.through
         self._forward_join()
@@ -232,7 +244,7 @@ class SqlJoin:
         self.own_alias = self.next_alias
         self.target_field = self.next_model.Meta.model_fields[self.relation_name]
 
-    def process_m2m_related_name_change(self, reverse: bool = False) -> str:
+    def _process_m2m_related_name_change(self, reverse: bool = False) -> str:
         """
         Extracts relation name to link join through the Through model declared on
         relation field.
@@ -272,24 +284,21 @@ class SqlJoin:
         Process order_by causes for non m2m relations.
 
         """
-        to_table = self.next_model.Meta.table.name
-        to_key, from_key = self.get_to_and_from_keys()
+        to_key, from_key = self._get_to_and_from_keys()
 
-        on_clause = self.on_clause(
+        on_clause = self._on_clause(
             previous_alias=self.own_alias,
             from_clause=f"{self.target_field.owner.Meta.tablename}.{from_key}",
-            to_clause=f"{to_table}.{to_key}",
+            to_clause=f"{self.to_table}.{to_key}",
+        )
+        target_table = self.alias_manager.prefixed_table_name(
+            self.next_alias, self.to_table
         )
-        target_table = self.alias_manager.prefixed_table_name(self.next_alias, to_table)
         self.select_from = sqlalchemy.sql.outerjoin(
             self.select_from, target_table, on_clause
         )
 
-        pkname_alias = self.next_model.get_column_alias(self.next_model.Meta.pkname)
-        if not self.target_field.is_multi:
-            self.get_order_bys(
-                to_table=to_table, pkname_alias=pkname_alias,
-            )
+        self._get_order_bys()
 
         # TODO: fix fields and exclusions for through model?
         self_related_fields = self.next_model.own_table_columns(
@@ -305,88 +314,35 @@ class SqlJoin:
         )
         self.used_aliases.append(self.next_alias)
 
-    def _replace_many_to_many_order_by_columns(self, part: str, new_part: str) -> None:
-        """
-        Substitutes the name of the relation with actual model name in m2m order bys.
-
-        :param part: name of the field with relation
-        :type part: str
-        :param new_part: name of the target model
-        :type new_part: str
-        """
-        if self.order_columns:
-            split_order_columns = [
-                x.split("__") for x in self.order_columns if "__" in x
-            ]
-            for condition in split_order_columns:
-                if self._check_if_condition_apply(condition, part):
-                    condition[-2] = condition[-2].replace(part, new_part)
-            self.order_columns = [x for x in self.order_columns if "__" not in x] + [
-                "__".join(x) for x in split_order_columns
-            ]
-
-    @staticmethod
-    def _check_if_condition_apply(condition: List, part: str) -> bool:
-        """
-        Checks filter conditions to find if they apply to current join.
-
-        :param condition: list of parts of condition split by '__'
-        :type condition: List[str]
-        :param part: name of the current relation join.
-        :type part: str
-        :return: result of the check
-        :rtype: bool
-        """
-        return len(condition) >= 2 and (
-            condition[-2] == part or condition[-2][1:] == part
+    def _set_default_primary_key_order_by(self) -> None:
+        clause = ormar.OrderAction(
+            order_str=self.next_model.Meta.pkname,
+            model_cls=self.next_model,
+            alias=self.next_alias,
         )
+        self.sorted_orders[clause] = clause.get_text_clause()
 
-    def set_aliased_order_by(self, condition: List[str], to_table: str,) -> None:
-        """
-        Substitute hyphens ('-') with descending order.
-        Construct actual sqlalchemy text clause using aliased table and column name.
-
-        :param condition: list of parts of a current condition split by '__'
-        :type condition: List[str]
-        :param to_table: target table
-        :type to_table: sqlalchemy.sql.elements.quoted_name
-        """
-        direction = f"{'desc' if condition[0][0] == '-' else ''}"
-        column_alias = self.next_model.get_column_alias(condition[-1])
-        order = text(f"{self.next_alias}_{to_table}.{column_alias} {direction}")
-        self.sorted_orders["__".join(condition)] = order
-
-    def get_order_bys(self, to_table: str, pkname_alias: str,) -> None:  # noqa: CCR001
+    def _get_order_bys(self) -> None:  # noqa: CCR001
         """
         Triggers construction of order bys if they are given.
         Otherwise by default each table is sorted by a primary key column asc.
-
-        :param to_table: target table
-        :type to_table: sqlalchemy.sql.elements.quoted_name
-        :param pkname_alias: alias of the primary key column
-        :type pkname_alias: str
         """
         alias = self.next_alias
         if self.order_columns:
             current_table_sorted = False
-            split_order_columns = [
-                x.split("__") for x in self.order_columns if "__" in x
-            ]
-            for condition in split_order_columns:
-                if self._check_if_condition_apply(condition, self.relation_name):
+            for condition in self.order_columns:
+                if condition.check_if_filter_apply(
+                    target_model=self.next_model, alias=alias
+                ):
                     current_table_sorted = True
-                    self.set_aliased_order_by(
-                        condition=condition, to_table=to_table,
-                    )
-            if not current_table_sorted:
-                order = text(f"{alias}_{to_table}.{pkname_alias}")
-                self.sorted_orders[f"{alias}.{pkname_alias}"] = order
+                    self.sorted_orders[condition] = condition.get_text_clause()
+            if not current_table_sorted and not self.target_field.is_multi:
+                self._set_default_primary_key_order_by()
 
-        else:
-            order = text(f"{alias}_{to_table}.{pkname_alias}")
-            self.sorted_orders[f"{alias}.{pkname_alias}"] = order
+        elif not self.target_field.is_multi:
+            self._set_default_primary_key_order_by()
 
-    def get_to_and_from_keys(self) -> Tuple[str, str]:
+    def _get_to_and_from_keys(self) -> Tuple[str, str]:
         """
         Based on the relation type, name of the relation and previous models and parts
         stored in JoinParameters it resolves the current to and from keys, which are
@@ -396,7 +352,7 @@ class SqlJoin:
         :rtype: Tuple[str, str]
         """
         if self.target_field.is_multi:
-            to_key = self.process_m2m_related_name_change(reverse=True)
+            to_key = self._process_m2m_related_name_change(reverse=True)
             from_key = self.main_model.get_column_alias(self.main_model.Meta.pkname)
 
         elif self.target_field.virtual:
diff --git a/ormar/queryset/prefetch_query.py b/ormar/queryset/prefetch_query.py
index 7abf4c6..08d2675 100644
--- a/ormar/queryset/prefetch_query.py
+++ b/ormar/queryset/prefetch_query.py
@@ -20,6 +20,7 @@ from ormar.queryset.utils import extract_models_to_dict_of_lists, translate_list
 if TYPE_CHECKING:  # pragma: no cover
     from ormar import Model
     from ormar.fields import ForeignKeyField, BaseField
+    from ormar.queryset import OrderAction
 
 
 def add_relation_field_to_fields(
@@ -128,7 +129,7 @@ class PrefetchQuery:
         exclude_fields: Optional[Union[Dict, Set]],
         prefetch_related: List,
         select_related: List,
-        orders_by: List,
+        orders_by: List["OrderAction"],
     ) -> None:
 
         self.model = model_cls
@@ -141,7 +142,9 @@ class PrefetchQuery:
         self.models: Dict = {}
         self.select_dict = translate_list_to_dict(self._select_related)
         self.orders_by = orders_by or []
-        self.order_dict = translate_list_to_dict(self.orders_by, is_order=True)
+        self.order_dict = translate_list_to_dict(
+            [x.query_str for x in self.orders_by], is_order=True
+        )
 
     async def prefetch_related(
         self, models: Sequence["Model"], rows: List
diff --git a/ormar/queryset/query.py b/ormar/queryset/query.py
index edb28c1..7c5a211 100644
--- a/ormar/queryset/query.py
+++ b/ormar/queryset/query.py
@@ -8,11 +8,12 @@ from sqlalchemy import text
 import ormar  # noqa I100
 from ormar.models.helpers.models import group_related_list
 from ormar.queryset import FilterQuery, LimitQuery, OffsetQuery, OrderQuery
-from ormar.queryset.filter_action import FilterAction
+from ormar.queryset.actions.filter_action import FilterAction
 from ormar.queryset.join import SqlJoin
 
 if TYPE_CHECKING:  # pragma no cover
     from ormar import Model
+    from ormar.queryset import OrderAction
 
 
 class Query:
@@ -26,7 +27,7 @@ class Query:
         offset: Optional[int],
         fields: Optional[Union[Dict, Set]],
         exclude_fields: Optional[Union[Dict, Set]],
-        order_bys: Optional[List],
+        order_bys: Optional[List["OrderAction"]],
         limit_raw_sql: bool,
     ) -> None:
         self.query_offset = offset
@@ -45,7 +46,7 @@ class Query:
         self.select_from: List[str] = []
         self.columns = [sqlalchemy.Column]
         self.order_columns = order_bys
-        self.sorted_orders: OrderedDict = OrderedDict()
+        self.sorted_orders: OrderedDict[OrderAction, text] = OrderedDict()
         self._init_sorted_orders()
 
         self.limit_raw_sql = limit_raw_sql
@@ -58,28 +59,6 @@ class Query:
             for clause in self.order_columns:
                 self.sorted_orders[clause] = None
 
-    @property
-    def prefixed_pk_name(self) -> str:
-        """
-        Shortcut for extracting prefixed with alias primary key column name from main
-        model
-        :return: alias of pk column prefix with table name.
-        :rtype: str
-        """
-        pkname_alias = self.model_cls.get_column_alias(self.model_cls.Meta.pkname)
-        return f"{self.table.name}.{pkname_alias}"
-
-    def alias(self, name: str) -> str:
-        """
-        Shortcut to extracting column alias from given master model.
-
-        :param name: name of column
-        :type name: str
-        :return: alias of given column name
-        :rtype: str
-        """
-        return self.model_cls.get_column_alias(name)
-
     def apply_order_bys_for_primary_model(self) -> None:  # noqa: CCR001
         """
         Applies order_by queries on main model when it's used as a subquery.
@@ -88,16 +67,13 @@ class Query:
         """
         if self.order_columns:
             for clause in self.order_columns:
-                if "__" not in clause:
-                    text_clause = (
-                        text(f"{self.table.name}.{self.alias(clause[1:])} desc")
-                        if clause.startswith("-")
-                        else text(f"{self.table.name}.{self.alias(clause)}")
-                    )
-                    self.sorted_orders[clause] = text_clause
+                if clause.is_source_model_order:
+                    self.sorted_orders[clause] = clause.get_text_clause()
         else:
-            order = text(self.prefixed_pk_name)
-            self.sorted_orders[self.prefixed_pk_name] = order
+            clause = ormar.OrderAction(
+                order_str=self.model_cls.Meta.pkname, model_cls=self.model_cls
+            )
+            self.sorted_orders[clause] = clause.get_text_clause()
 
     def _pagination_query_required(self) -> bool:
         """
@@ -208,7 +184,9 @@ class Query:
             for filter_clause in self.exclude_clauses
             if filter_clause.table_prefix == ""
         ]
-        sorts_to_use = {k: v for k, v in self.sorted_orders.items() if "__" not in k}
+        sorts_to_use = {
+            k: v for k, v in self.sorted_orders.items() if k.is_source_model_order
+        }
         expr = FilterQuery(filter_clauses=filters_to_use).apply(expr)
         expr = FilterQuery(filter_clauses=excludes_to_use, exclude=True).apply(expr)
         expr = OrderQuery(sorted_orders=sorts_to_use).apply(expr)
diff --git a/ormar/queryset/queryset.py b/ormar/queryset/queryset.py
index ba55586..2202abd 100644
--- a/ormar/queryset/queryset.py
+++ b/ormar/queryset/queryset.py
@@ -21,6 +21,7 @@ import ormar  # noqa I100
 from ormar import MultipleMatches, NoMatch
 from ormar.exceptions import ModelError, ModelPersistenceError, QueryDefinitionError
 from ormar.queryset import FilterQuery
+from ormar.queryset.actions.order_action import OrderAction
 from ormar.queryset.clause import QueryClause
 from ormar.queryset.prefetch_query import PrefetchQuery
 from ormar.queryset.query import Query
@@ -514,7 +515,12 @@ class QuerySet(Generic[T]):
         if not isinstance(columns, list):
             columns = [columns]
 
-        order_bys = self.order_bys + [x for x in columns if x not in self.order_bys]
+        orders_by = [
+            OrderAction(order_str=x, model_cls=self.model_cls)  # type: ignore
+            for x in columns
+        ]
+
+        order_bys = self.order_bys + [x for x in orders_by if x not in self.order_bys]
         return self.__class__(
             model_cls=self.model,
             filter_clauses=self.filter_clauses,
@@ -713,7 +719,14 @@ class QuerySet(Generic[T]):
             return await self.filter(**kwargs).first()
 
         expr = self.build_select_expression(
-            limit=1, order_bys=[f"{self.model.Meta.pkname}"] + self.order_bys
+            limit=1,
+            order_bys=[
+                OrderAction(
+                    order_str=f"{self.model.Meta.pkname}",
+                    model_cls=self.model_cls,  # type: ignore
+                )
+            ]
+            + self.order_bys,
         )
         rows = await self.database.fetch_all(expr)
         processed_rows = self._process_query_result_rows(rows)
@@ -742,7 +755,14 @@ class QuerySet(Generic[T]):
 
         if not self.filter_clauses:
             expr = self.build_select_expression(
-                limit=1, order_bys=[f"-{self.model.Meta.pkname}"] + self.order_bys
+                limit=1,
+                order_bys=[
+                    OrderAction(
+                        order_str=f"-{self.model.Meta.pkname}",
+                        model_cls=self.model_cls,  # type: ignore
+                    )
+                ]
+                + self.order_bys,
             )
         else:
             expr = self.build_select_expression()
diff --git a/ormar/queryset/utils.py b/ormar/queryset/utils.py
index 6b98028..9445fe0 100644
--- a/ormar/queryset/utils.py
+++ b/ormar/queryset/utils.py
@@ -232,16 +232,24 @@ def get_relationship_alias_model_and_str(
     is_through = False
     model_cls = source_model
     previous_model = model_cls
+    previous_models = [model_cls]
     manager = model_cls.Meta.alias_manager
     for relation in related_parts[:]:
         related_field = model_cls.Meta.model_fields[relation]
+
         if related_field.is_through:
+            # through is always last - cannot go further
             is_through = True
-            related_parts = [
-                x.replace(relation, related_field.related_name) if x == relation else x
-                for x in related_parts
+            related_parts.remove(relation)
+            through_field = related_field.owner.Meta.model_fields[
+                related_field.related_name or ""
             ]
-            relation = related_field.related_name
+            if len(previous_models) > 1 and previous_models[-2] == through_field.to:
+                previous_model = through_field.to
+                relation = through_field.related_name
+            else:
+                relation = related_field.related_name
+
         if related_field.is_multi:
             previous_model = related_field.through
             relation = related_field.default_target_field_name()  # type: ignore
@@ -250,6 +258,8 @@ def get_relationship_alias_model_and_str(
         )
         model_cls = related_field.to
         previous_model = model_cls
+        if not is_through:
+            previous_models.append(previous_model)
     relation_str = "__".join(related_parts)
 
     return table_prefix, model_cls, relation_str, is_through
diff --git a/tests/test_m2m_through_fields.py b/tests/test_m2m_through_fields.py
index 9d7f38a..5c1b44f 100644
--- a/tests/test_m2m_through_fields.py
+++ b/tests/test_m2m_through_fields.py
@@ -34,6 +34,14 @@ class PostCategory(ormar.Model):
     param_name: str = ormar.String(default="Name", max_length=200)
 
 
+class Blog(ormar.Model):
+    class Meta(BaseMeta):
+        pass
+
+    id: int = ormar.Integer(primary_key=True)
+    title: str = ormar.String(max_length=200)
+
+
 class Post(ormar.Model):
     class Meta(BaseMeta):
         pass
@@ -41,6 +49,7 @@ class Post(ormar.Model):
     id: int = ormar.Integer(primary_key=True)
     title: str = ormar.String(max_length=200)
     categories = ormar.ManyToMany(Category, through=PostCategory)
+    blog = ormar.ForeignKey(Blog)
 
 
 @pytest.fixture(autouse=True, scope="module")
@@ -146,18 +155,86 @@ async def test_filtering_by_through_model() -> Any:
         )
 
         post2 = (
-            await Post.objects.filter(postcategory__sort_order__gt=1)
-                .select_related("categories")
-                .get()
+            await Post.objects.select_related("categories")
+            .filter(postcategory__sort_order__gt=1)
+            .get()
         )
         assert len(post2.categories) == 1
         assert post2.categories[0].postcategory.sort_order == 2
 
         post3 = await Post.objects.filter(
-            categories__postcategory__param_name="volume").get()
+            categories__postcategory__param_name="volume"
+        ).get()
         assert len(post3.categories) == 1
         assert post3.categories[0].postcategory.param_name == "volume"
 
+
+@pytest.mark.asyncio
+async def test_deep_filtering_by_through_model() -> Any:
+    async with database:
+        blog = await Blog(title="My Blog").save()
+        post = await Post(title="Test post", blog=blog).save()
+
+        await post.categories.create(
+            name="Test category1",
+            postcategory={"sort_order": 1, "param_name": "volume"},
+        )
+        await post.categories.create(
+            name="Test category2", postcategory={"sort_order": 2, "param_name": "area"}
+        )
+
+        blog2 = (
+            await Blog.objects.select_related("posts__categories")
+            .filter(posts__postcategory__sort_order__gt=1)
+            .get()
+        )
+        assert len(blog2.posts) == 1
+        assert len(blog2.posts[0].categories) == 1
+        assert blog2.posts[0].categories[0].postcategory.sort_order == 2
+
+        blog3 = await Blog.objects.filter(
+            posts__categories__postcategory__param_name="volume"
+        ).get()
+        assert len(blog3.posts) == 1
+        assert len(blog3.posts[0].categories) == 1
+        assert blog3.posts[0].categories[0].postcategory.param_name == "volume"
+
+
+@pytest.mark.asyncio
+async def test_ordering_by_through_model() -> Any:
+    async with database:
+        post = await Post(title="Test post").save()
+        await post.categories.create(
+            name="Test category1",
+            postcategory={"sort_order": 2, "param_name": "volume"},
+        )
+        await post.categories.create(
+            name="Test category2", postcategory={"sort_order": 1, "param_name": "area"}
+        )
+        await post.categories.create(
+            name="Test category3",
+            postcategory={"sort_order": 3, "param_name": "velocity"},
+        )
+
+        post2 = (
+            await Post.objects.select_related("categories")
+            .order_by("-postcategory__sort_order")
+            .get()
+        )
+        assert len(post2.categories) == 3
+        assert post2.categories[0].name == "Test category3"
+        assert post2.categories[2].name == "Test category2"
+
+        post3 = (
+            await Post.objects.select_related("categories")
+            .order_by("categories__postcategory__param_name")
+            .get()
+        )
+        assert len(post3.categories) == 3
+        assert post3.categories[0].postcategory.param_name == "area"
+        assert post3.categories[2].postcategory.param_name == "volume"
+
+
 # TODO: check/ modify following
 
 # add to fields with class lower name (V)
@@ -166,10 +243,12 @@ async def test_filtering_by_through_model() -> Any:
 # creating in queryset proxy (dict with through name and kwargs) (V)
 # loading the data into model instance of though model (V) <- fix fields ane exclude
 # accessing from instance (V) <- no both sides only nested one is relevant, fix one side
-# filtering in filter (through name normally) (V) < - table prefix from normal relation, check if is_through needed
+# filtering in filter (through name normally) (V) < - table prefix from normal relation,
+#               check if is_through needed, resolved side of relation
+# ordering by in order_by
+
 
 # updating in query
-# ordering by in order_by
 # modifying from instance (both sides?)
 # including/excluding in fields?
 # allowing to change fk fields names in through model?