An example will (hopefully) make it clearer.
I have three images red.png, blue.png, green.png:



and I want to combine them to give this result:

The yellow borders are not part of the images, but are there to indicate the dimensions of each of the images (note that the output image does not extend all the way to the edges). Note also that the input image relative sizes are in differing proportions to their sizes in the output image.
So in this example, my requirements would be:
- Output image size is to be 100x100
Red.png is to be sized so that it is centred (i.e. so that the centre of the input image is positioned at +50, +50 in the output) and occupies 70 pixels (i.e. 15 pixels clear to the border)
Blue.png is to be positioned so the centre is at +70, +70, and sized so that the circle is 43 pixels in diameter.
Green.png is to be positioned so the centre is at +70, +45, and sized so that the circle is 35 pixels in diameter.
- keeping the sizing of each of the input component layers from interacting with each other.
forcing the output image to be a particular fixed size.
dealing with arbitrary input image sizes