There is nothing “simple” about what you want to do. Appearances are deceiving.
As far as the problem at hand there is no right or wrong answer.
The likely best solution is to remove the head from the character, assemble it on the skeletal mesh at runtime, and set it to not render but cast shadow. You can then control what it does on reflections as well.
You can couple this with a negative value Spring (camera boom).
Attach the spring to the head bone. Set the value to -30.
Attach the camera to the spring’s socket.
Set the camera to detect collision / change the size of the test to match the approx size of the face.
That will help when the character tries to shove his face into walls.
Play with the spring value to get a good starting distance. 30 may be excessive.
Short of decoupling the head your only other option is to offset either the camera or the spring arm manually by however many units gets you past the geometry that is rendering inside the head (the model normals point that way).