Touch-screens are becoming increasingly ubiquitous. They have great appeal due to their capabilities to support new forms of human interaction, including their abilities to interpret rich gestural inputs, render flexible user interfaces and enable multi-user interactions. However, the technology creates new challenges and barriers for users with limited levels of vision and motor abilities. The PhD work described in this paper proposes a technique combining Shared User Models (SUM) and adaptive interfaces to improve the accessibility of touch-screen devices for people with low levels of vision and motor ability. SUM, built from an individual's interaction data across multiple applications and devices, is used to infer new knowledge of their abilities and characteristics, without the need for continuous calibration exercises or user configurations. This approach has been realized through the development of an open source software framework to support the creation of applications that make use of SUM to adapt interfaces that match the needs of individual users.