Recent works suggested the use of self-organising spatial patterns, enacted as computational fields, to the problem of steering users towards their desired destination in complex environments. In a rich and open scenarios, various contextual services can enter the system providing additional information that can be exploited to guide users by suggesting paths that more likely satisfy their preferences. This can be done composing new information with the steering services already available in the system. Since the type and number of new services available can vary over time, such compositions must be identified dynamically, in an autonomous, spontaneous and unsupervised way.
Moreover, in order to avoid the system to be overflooded by services that do not match any user preferences, a mechanism for identifying useless compositions and for removing them should be provided. In this paper we investigate this problem and propose a \emph{self}-composition approach envisioned to support the composition of new services together with basic services for crowd steering, such that, depending on the actual (spatial/temporal) context in which the composition is deployed, users are steered across the path that better fit their preferences.