Quality of Service (QoS) methods aim at trading quality against resource requirements to meet the constraints dictated by the application functionality and the execution platform. QoS is relevant in multimedia tasks since these applications are typically scalable systems. To exploit the scalability property for improving quality, a reliable model of the relation between scalable parameters and quality/resources is required. The traditional QoS approach requires a deep knowledge of the execution platform and a reasonably accurate prediction of the expected configurations. This paper proposes an alternative black-box data analysis approach. The advantage is that it requires no a priori assumptions about the correlation between quality/resources and parameters and it can easily adapt to situations of high complexity, changing platforms and heterogeneous environments. Some preliminary experiments with the QoS modelling of the Visual Texture Coding (VTC) functionality of a MPEG-4 decoder using a local learning technique are presented to support the claim.