Real life optimization often concerns difficult objective functions, in two aspects, namely that gradients are unavailable, and that evaluation of the objective function takes a long time. Such problems are often attacked with model building algorithms, where an approximation of the function is constructed and solved, in order to find a new promising point to evaluate. We study several ways of saving time by using parallel calculations in the context of model building algorithms, which is not trivial, since such algorithms are inherently sequential. We present a number of ideas that has been implemented and tested on a large number of known test functions, and a few new ones. The computational results reveal that some ideas are quite promising.