It is not entirely clear from your post if you are talking about a categorical predictor (one of the variables in X) or the categorical response (Y, or class labels). I will assume the latter.
classregtree can only predict into one of the classes included in the training data. How would you determine if an observation is not one of the three categories? If you can do it, classregtree can do it too, as long as you have enough data for this 4th category and use these data to grow the tree.
