The sprites
are implemented as 3D plains as I understant it.
However they're locked to their own rendering pipeline and you have less control over them. I also believe they are a bit slower (no idea why) than emulating sprites with 3d plains facing the camera yourself. Your biggest problem would be positioning them at an appropriate distance from the camera to make them have their correct pixel sizes on screen.
As for showing / hiding the plains, I'm not sure, it might be referring to how you could have your "real" (3D) camera fixed in place and implement your own 2D camera system by shuffling all sprites positions and then hiding the ones that end up outside the screen to speed things up (although excluding would be more efficient for that).
"Why do programmers get Halloween and Christmas mixed up?" Because Oct(31) = Dec(25)