The aim of this paper is to address the dial-a-ride problem with heterogeneous users in which the vehicle capacity can be modified en-route by reconfiguring its internal layout. The work is motivated by the daily transport of children with disabilities performed by a private company based in Lyon Métropole, France. Every day, a fleet of configurable vehicles is available to transport children to medico-social establishments. The objective of this work is then to help route planners with the fleet dimensioning and take reconfiguration opportunities into consideration in the design of routes. Due to the number of passengers and vehicles, real-size instances are intractable for mix-integer programing solvers and exact solution methods. Thus, a large neighborhood search meta-heuristic combined with a set covering component is proposed. The resulting framework is evaluated on real life instances from the transport company.