Seems like there is not a lot of information available on Model Targets other than you can do it. So i wanted to start a thread for sharing best practices. So far my only problems have been on the detection side in View. The only guidelines i have found on improving detection are in the Vuforia API documentation on this feature, specifically in the instructions for the Vuforia target generator app.
For stable detection, detection positions with a fronto-parallel view onto the object should be avoided. If possible, choose a position with a diagonal view, such that two or three sides of the object can be seen. Additionally, detection positions with many parallel lines should be avoided. For smaller objects, detection positions where the object only fills a small part of the image should be avoided.
I had to look up "fronto-parallel" and could only find machine vision papers using it. Basically don't use a view with a big flat surface parallel to the camera view plane. If your model is a box use any view except on straight on to a side.