There don't seem to be many universally recognised patterns for 3D gestures, such as you might use to interact with something like the Kinect. A previous question looks at Kinect gestures, but there aren't too many satisfactory answers.
Given that, it seems like developers have to work from scratch on gestures - this isn't ideal, and could be solved by creating a decent pattern library, like this one for multi-touch gestures.
The question, therefore, is this: what methods can be used to create such a library? What principles should be followed, and are there any common techniques for establishing patterns like this (patterns for creating patterns, if you will)? And are there any considerations unique to 3D gestures?