You should know, that animation have often multiple layers of skeletons. A common way is to define a skeleton for the deformation of the skin (used in-game) and a layer of drivers/targets (and a layer for rag-dolls etc.). Both layers are then connected by constrains. The trick is, that you will only animate the drivers which on the other hand will animate the real deformation skeleton through the constrains. That has several advantages, the most import is, that you can modify the underlying deformation skeleton without loosing your animation, because the driver skeleton does not change.
You can utilize this to some degree when importing motion capture data, try something like this:
1. Build your deformation skeleton and bind it to the skin.
2. Import your motion capture data.
3. Add constrains, so that your deformation skeleton will follow the motrion capture data. You don't need to map all bones, a very simple startup would be to use the feets, hands, head and torso, add more targets to refine the quality.
4. Tweak the animation by manipulating the drivers, not the deformation skeleton !
Thought, defining working constrains is an art itself, a reason that you find artists in game studios with the sole task to create rigs, so it will not be an easy task. Best to use motion capture data which could be mapped more or less 1:1 on your skeleton.