Python's flexible, duck-typed object system lowers the cost of architectural options that are more difficult to exercise in more rigid languages (yes, we are thinking of C++). One of these is carefully separating your data model (the classes and data structures that represent whatever state your application is designed to manipulate) from your controller (the classes that implement your user interface.
In Python, a design pattern that frequently applies is to have one master editor/controller class that encapsulates your user interface (with, possibly, small helper classes for stateful widgets) and one master model class that encapsulates your application state (probably with some members that are themselves instances of small data-representation classes). The controller calls methods in the model to do all its data manipulation; the model delegates screen-painting and input-event processing to the controller.
Narrowing the interface between model and controller makes it easier to avoid being locked into early decisions about either part by adhesions with the other one. It also makes downstream maintainance and bug diagnosis easier.