Maple is an incredibly powerful tool for direct and inverse kinematics of simple and very complex mechanisms as well for their equations of motion derivation. I'm using it for these purposes for many years and it never let me down.
Regarding the use of a manipulators as machine tool, from a practical point of view, you may encounter some drawbacks related to uncertainty. Generally, any sensor or any manipulator part is affected by some amount of uncertainity (in its reading/real length/sensor mounting and positioning etc.) and as consequence the final position of the end effector cannot be determined exactly. On top of that, the error in positioning may change a lot accordingly to the "way" you reach a certain point in space.
As example (maple worksheet attached), let us assume that the angular readings of a simple planar robot ( and ) and its arms lenghts ( and ) are affected by some amount error (that can be modeled with a normal distribution). The position of the end effector, using recursive kinematics, is given by:
error -> (the same for , and )
As you can see (a bit exaggerate to be visible), the uncertainity in positioning changes a lot in shape as function of the arms angle.
Here you reach the same point in space but using different arms angles. As you can see the end effector position uncertainity cloud shape and orientation changes.
Maple worksheet -> manipulator.mw