I'm thinking of something like this:
var sphere = new Sphere(radius);
var square = new Square(length);
sphere.insert(square, x, y, z).subtract();
scene.insert(sphere, x, y, z);
Basically each model would be like an object. More advanced models could be made by combining other models, specifying how the models should interact, and then once done the finished product could be simplified to just the remaining points rather than the formula (basically compressed?).
I know there's POV-Ray for the scripting aspect. From what I can tell, though, the rendering isn't done until specified on the command line.
Are there any other options out there? I'd like to compare several at once if possible.