The open architecture of the Internet has enabled its massive growth and success by facilitating easy connectivity between hosts. At the same time, the Internet has also opened itself up to abuse, e.g. arising out of unsolicited communication, both intentional and unintentional. It remains an open question as to how best servers should protect themselves from malicious clients whilst offering good service to innocent clients. There has been research on behavioural profiling and reputation of clients, mostly at the network level and also for email as an application, to detect malicious clients. However, this area continues to pose open research challenges. This thesis is motivated by the need for a generalised framework capable of aiding efficient detection of malicious clients while being able to reward clients with behaviour profiles conforming to the acceptable use and other relevant policies. The main contribution of this thesis is a novel, generalised, context-aware, policy independent, privacy preserving framework for developing and sharing client reputation based on behavioural history. The framework, augmenting existing protocols, allows fitting in of policies at various stages, thus keeping itself open and flexible to implementation. Locally recorded behavioural history of clients with known identities are translated to client reputations, which are then shared globally. The reputations enable privacy for clients by not exposing the details of their behaviour during interactions with the servers. The local and globally shared reputations facilitate servers in selecting service levels, including restricting access to malicious clients. We present results and analyses of simulations, with synthetic data and some proposed example policies, of client-server interactions and of attacks on our model. Suggestions presented for possible future extensions are drawn from our experiences with simulation.