Realtime musical expression through synthesis is notoriously difficult. The complete potential of a sound engine is traditionally available only at design time. In this paper two mapping strategies are presented, adressing this problem. Based on dynamic and random many-to-many mappings between control space and synthesis space, they erase the line between sound editing and synthesizer performance, with an emphasis on free improvisation. One strategy is based on the addition of random vectors in synthesis space, weighted by control parameters. The other is based on a gravity analogy, interpolating between random points in synthesis space, with the gravities controlled by player interaction. Vectors and point sets can be scaled and shifted during performance, allowing dynamic exploration of vast soundspaces or minute timberal control. The mappings have been adopted to a wide range of musical interfaces. Together with suitable sound engines, surprisingly expressive performance instruments have been created, and they have been used in regular rehearsals, concerts and recording sessions over the last two years.