Most computational techniques for the design of combinatorial libraries have concentrated solely on maximizing the diversity of the selected subset or its similarity to a known target. However, such libraries can produce high-throughput screening hits with properties that make them unsuitable to take forward into medicinal chemistry. This article describes software that allows the design of library subsets to simultaneously optimize a library's diversity or similarity to a target, properties (such as drug likeness) of the library members, properties (such as cost) of the reagents required to make them, and efficiency of synthesis in arrays or mixtures. Example are given showing that libraries can be designed to contain drug-like molecules with only a small trade-off in terms of the maximum possible diversity, and that the cost of the library, in terms of the reagents required to make it, can be contained. Other examples show that libraries can be designed to minimize the deconvolution problem or to maximize the number of molecules predicted to be active while also being designed for efficiency of synthesis.