Classification of Classes and Subclasses

I am playing around with the Classification Learner to classify my data and am looking for a general advise how to deal with classification.

I have two big classes: Cars and airplanes. The 'car' class can be split into several subclasses, e.g. 'GM', 'Ford', 'Toyota'. The same holds true for the 'airplane' class (e.g. 'Boeing', 'Airbus').

When training models with only the 'car' or 'airplane' information, the accuracy is very high. There are almost no cars identified as airplanes and vice versa. However, when I train directly with the subclass information (no 'cars' and 'airplanes', but only 'GM', 'Ford', 'Boeing' etc.), there are some 'Boeing' objects identified as 'Ford'.

What's the best approach to tackle this? My only idea is to create a model that first decides whether an object is a 'car' or 'airplane' and then have sole models for each subclass. So that only objects identified as 'car' are run through a Ford-GM-Toyota-Model and only 'airplane' through a Airbus-Boeing-Model.

On the other hand this seems to be a bit too clunky, hence I look for some ideas.