Decentralized systems are scalable by design but also difficult to coordinate due to their weak coupling. Replicating data in these geo-distributed systems is therefore a challenge inherent to their structure. The two contributions of this thesis exploit the heterogeneity of user requirements and enable personalizable quality of services for data replication in decentralized systems. Our first contribution Gossip Primary-Secondary enables the consistency criterion Update consistency Primary-Secondary to offer differentiated guarantees in terms of consistency and message delivery latency for large-scale data replication. Our second contribution Dietcoin enriches Bitcoin with diet nodes that can (i) verify the correctness of entire subchains of blocks while avoiding the exorbitant cost of bootstrap verification and (ii) personalize their own security and resource consumption guarantees.