Yes, it is possible, but it is not one of the out-of-the-box modules yet. So you would have to find the method you like best. Maybe start with opencv 3.3 and python. This way, you can directly port that code to JeVois without the need to install cross-compilers, etc. Have a look here for an example:
https://www.superdatascience.com/opencv-face-recognition/
and there are many more online. Other libraries are available, possibly with better performance, but require C++ programming, for example dlib. As far as performance, it will depend both on th emethod and on your training data.