Well something like this... The first pass renderst the model into the depth buffer so the front most polygons define the depth buffer. The second pass uses the depth buffer from the first pass rendering the model with the red shape, but only the frontside polygons.
