In the last two decades the need for ever higher resolution simulations of atmospheric flow has required the distribution of the problem over networks of parallel processors. While methods, e.g. finite differences, finite volumes or finite elements, to solve these grid-based problems vary, the difficulties in the parallelization are similar. This paper investigates the commonness of parallel atmospheric flow applications and suggests basic functions to parallelize a wider range. An implementation of these functions as a reusable library is presented, as well as parallelization results of several key components of general circulation models.