In our work on computational design of expressive gestural interaction, we experienced various challenges for designing accurate methods for the task at end, and their under-standability and usability from the user perspective. In this proposal we present the computational approach developed in our research relying on volitional, personal, variations of gesture execution. In prior work, gesture variations have often been considered as undesirable. We present the modelling strategy undertaken to tackle this challenge and present two examples of models developed. By this proposal, we aim to discuss the encountered challenges for machine learning and human-computer interaction.