If you define the radius of a beam from the standard deviation of its profile (up to a constant factor), then for a given waist radius, gaussian beams are the ones that diverge the less (at infinity).
In paraxial optics, with gauss approximation etc., a perfect spherical lens or mirror will always transform a gaussian beam into another gaussian beam, hence a beam that has minimal divergence for a given waist radius. So, in an ideal case, starting from a diffraction-limited beam (=gaussian), the beam will always be diffraction-limited because it will stay gaussian.
I started writing a detailed answer on how to collimate beams, but I got lost into details and I'm not entirely sure what you want to do. My advice would be to play with a nice and simple gaussian beam software like this http://sourceforge.net/projects/gaussianbeam/ to see what optical components you need. Hint: if you want a beam with a very small divergence, you need a beam with a large diameter to start with, so you might have to expand your beam before collimating it.
Fresnel beams may allow you to have a beam that is better collimated around a focal point than a gaussian beam, for a given waist radius. However, their divergences (defined at infinity) are always worse than those of gaussian beams.