In the last few years, the use of mobile devices in computational grids has seen a growing interest. Nevertheless, a more challenging issue, the dynamic establishment of mobile grids on wireless ad hoc networks, has been so far only partially addressed. The first contribution of this thesis is the proposal of a software architecture for mobile grids that can be used for both infrastructured and ad hoc wireless networks. In the execution of conventional applications in grids, the responsibility to provide the service is shared among the most resourceful mobile devices. In mobile grids, it is fundamental that resource discovery and selection of resources are jointly handled. This calls for a mechanism that promotes the automatic selection of the best resource providers amongst the discovered nodes, taking into account the requirements of the application. Discovery and selection, however, have been traditionally handled separately and in most approaches the selection of resources and services requires explicit intervention by the user of the mobile grid. As a second contribution of this thesis, we propose a protocol that integrates the phases of resource discovery and automatic selection in mobile grids, allowing that computational resource provisioning is scheduled among the most resourceful nodes. Due to the dynamics of the resources needed in a mobile grid (for example, free CPU time and available memory), the protocol is based solely on demand- driven broadcasts. However, mainly in multihop ad hoc wireless networks, this strategy can incur in overhead at the involved devices, due to the diffusion of requests and replies. A third contribution of this thesis is the development of a mechanism that allows to reduce this overhead by means of the suppression of redundant replies in the network. The mechanism has been implemented in the context of the proposal protocol, but can be applied as well to other query-based discovery protocols based on broadcasts. The experimental results obtained from executions in a testbed and through simulations show that the proposed protocol provides efficient load balancing between devices with an increasing number of requests. Moreover, it can be observed that the mechanism for suppression of replies scales well with respect to an increasing number of devices when compared to other discovery protocols in wireless ad hoc networks that are purely based on requests via broadcast.