Spontaneous wireless networks constructed out of mobile end-user devices (e.g. smartphones or tablets) are currently receiving considerable interest as they enable a wide range of novel, highly pervasive and user-centric network services and applications. In this paper, we focus on emergency-related scenarios, and we investigate the potential of spontaneous networks for providing Internet connectivity over the emergency area through the sharing of resources owned by the end-user devices. Novel and extremely flexible network deployment strategies are required in order to cope with the user mobility, the limited communication capabilities of wireless devices, and the intrinsic dynamics of traffic loads and QoS requirements. To this purpose, we propose here a novel approach toward the deployment of spontaneous networks composed by a new generation of wireless devices – called Stem Nodes (SNs) – to emphasize their ability to cover multiple network roles (e.g. gateway, router). The self-organization of the spontaneous network is then achieved through the local reconfiguration of each SN. Two complementary research contributions are provided. First, we describe the software architecture of a SN (which can be implemented on top of existing end-user devices), and we detail how a SN can manage its role set, eventually extending it through cooperation with other SNs. Second, we propose distributed algorithms, based on swarm intelligence principles, through which each SN can autonomously select its role, and self-elect to gateway or router, so that end-to-end performance are maximized while the lifetime of the spontaneous emergency network is prolonged. The ability of the proposed algorithm to guarantee adaptive and self-organizing network behaviors is demonstrated through extensive Omnet++ simulations, and through a prototype implementation of the SN architecture on a real testbed.