This is always a difficult thing for people to get their heads around. Instead of letting the controller worry about the actual buttons, it should be worried about what the view is allowed to do (ie the actions it can perform), which (presumably updates the model).
So, your view would actually handle the buttons events internally, but, instead of changing the state itself, it would notify the controller the a particular state has changed or action has been performed.
This communication would be managed via a series of interface contracts. This means that are particular controller is expecting to control a particular type of of view, but neither care about the actual implementation, so long as the contract between the two is maintained
With this in mind, it then means that your view can do what ever it likes and generate the "events" in anyway it likes, so long as the contract is upheld and you're not exposing parts of your view to other parts of the program which has no reason to reference it