Let's have a button Command property bound to a custom command.
When should I implement ICommand and when derive from RoutedCommand? I see that RoutedCommand implements ICommand.
In which case could I need to iplement an ICommand?
What about MVVM model? Which one suits better for this purpose?
As you have noticed the RoutedCommand class is an implementation of the ICommand interface, its main distinction if that its function is similar to that of a RoutedEvent:
The Execute and CanExecute methods on a RoutedCommand do not contain the application logic for the command as is the case with a typical ICommand, but rather, these methods raise events that traverse the element tree looking for an object with a CommandBinding. The event handlers attached to the CommandBinding contain the command logic.
The Execute method raises the PreviewExecuted and Executed events. The CanExecute method raises the PreviewCanExecute and CanExecute events.
In a case when you don't want the behavior of the RoutedCommand you'll be looking at your own implementation of ICommand. As for the MVVM pattern I can't say that one solution, it seems that everyone has their own methodology. However, here are a few approaches to this problem that I've come across:
The only thing I would add to Rich McGuire's answer is that RoutedCommands (and their more prevalent descendant RoutedUICommand have to be wired up with event handlers to work correctly.
Most MVVM implementations I have come across attempt to leverage binding against the ViewModel and thus the ViewModel (and not the View) owns the CanExecute/Execute logic.
In contrast, the event handlers move that burden to the View. The handling can then be propagated to the ViewModel, but this means a slightly higher degree of coupling between ViewModel and View (casting + method call, etc.).