It really depends on your application. Depth camera itself already gives you back the 3D information, so you probably don't use the pair as a conventional stereo system. In another word, you don't need to do stereo calibration. If you want to use the conventional checkerboard to do calibration for each camera (mainly for radial distortion correction), it is fine. But you cannot see the checkers clearly in depth camera. you should use the IR camera (if you are using the old version of kinect).
The calibration involved in kinect is mostly about the alignment between color and depth camera. Sometimes people also calibrate the depth camera to improve the depth accuracy. Both steps do not exactly follow the traditional checkerboard workflow. Also MS SDK already provides the alignment function. If you are able to use the latest computer vision system toolbox 14b from matlab, you can easily get the aligned frame, called alignColorToDepth. You can also get the point cloud from depth image, and start from there. For many applications, these are good enough to start post-process. For a customized calibration procedure, check the following research work: http://www.ee.oulu.fi/~dherrera/papers/2012-KinectJournal.pdf
