Not exactly a clear thought, but a thought none the less. That thought is having pure abstract base classes where the abstract methods retrieve data. The physics engine contains all the physics "knowledge" which includes when it needs information, but not necessarily how to get that information. You might then provide helper classes that actually store the data. So perhaps the physics engine uses the bounding information as though it is a tree. You have abstract methods that are appropriate for navigating any tree, i.e. first, previous, next, last child, parent, etc. You then provide a class for actually storing the heirarchy the makes for a minimum of code when used. All the user has to do is make sure those functions return the right value. So you have a bounding sphere data type that isn''t optional because that is what the method returns. The unknown part is where the model is and how it is stored. The user has to provide the ability to actually find the model. If they use your helper class then that is about all they have to do. Otherwise they may have to translate data types and do a bunch of other things. The point being that it is simply if they want it to be, but flexible if they need it to be.