The window polygon does not have to be "away" from the wall. There are some geometry considerations that depend on whether you are modeling your walls with thickness or as infinitely thin walls as in an energy model. What does need to be separated from the wall/window plane is the polygon you will use for the BSDF, if you are using one. In this case, you use glow or (light) as the material because with the 3-phase method, the window polygon is acting as a light source to the points in the space (or the pixels of a given view) for the view matrix.

It sounds like you may be conflating a simple(r) two-phase approach -- where you just create a single view matrix and use that for an annual climate based daylight simulation, and the 3-phase --where you have some dynamic element (shade, blind, louver/film). In the former, you would use a glass material, and calculate a view matrix; the latter, you'd have a glass material for your window, and a second polygon inside the window plane to accept a BSDF. This second polygon is what would have the glow or light material for the view matrix, and acts as the sender polygon for the daylight matrix.

I recommend you review Sarith's excellent tutorial here for more information on this.