Even with deep data, you won't be able to acheive this. You will be able to add objects to your scene after the rendering process (in nuke for example). The added object will be occluded by the previously rendered ones, according to his distance from camera. Deep data are used to store world position for each pixel composing your image. So if you get a glass in front of any object, you won't be able to get the position of the objects behind that glass. 1 pixel = 1 position