For purposes of speed, collision detection between two objects generally terminates once a single point of intersection has been found. Since you can't guarantee that the first detected point was the only point of collision -- or even representative of the average -- the locality of this point is of little use.
Really, you'll have to handle this on a case by case basis. It's easier to approximate a reasonable point of collision between simple shapes than complex ones, and different mathematical models will be useful for different combinations of shapes.
In the case of a ball hitting a wall, you can use the radius of the ball and its angle of motion to approximate the point of collision using some basic trigonometry.