@lekonna said:
However the model should be always solid so it would be enough for me to just check the normals of the faces and compare those to the camera and if they are not opposite i could just reverse the face.
If you imagine a box with 6 faces which are all aligned correctly, the 3 faces which are visible would have their normals match the camera and the three faces which are invisible would have their normals the opposite of the camera. So the "trick" is to determine which faces are visible, and which faces are obstructed by other faces.
Todd Burch was working of a script to solve this problem, but I never heard if he got it finished.
I don't know any easy way to determine if a face is visible or obscured in a particular view.