I'll talk as if it is to be attached to a live person:
I'll assume that the 'person'/clavicle is situated on the negative side of x, leaving a mouse-roller x-setup on the positive x side. The y-setup has no constraints (can be on both the + and - side). The z-setup will have to point up (negative z direction). They will all have to be attached to the clavicle (negative x side) giving the 'external structure' of the whole setup a rather massive appearance (with simple, but cumbersome mechanics) .. compared to an insignificant 3-axis gyro or acceleration-sensor.
[edit]
ehmm .. three axis fixed on the humeral head pointing strictly at (0,0,0), but the dials/wheels will have to be both aligned to these (collectively dynamic) axis and fixed to the clavicle. .. I think that you are heading into some professional fine-mechanical engineering. Or do I just lack imagination?