Non-functional requirements specification for an application and definition of required resources for its deployment are activities commonly performed at ad hoc, which can cause quality of service degradation due to inefficient use of re-sources. This problem is even greater with applications formed by the composition of multiple services. Cloud com-puting emerges as an alternative to provision of the needed resources, which is usually done using local infrastructure in conjunction with external providers. However, efficient resource management mechanisms and policies for offering quality services in these environments are necessary. This work is a PhD proposal, where is presented a middleware-based approach for cloud resources management and their provisioning according with non-funcional requirements and with greater usability by the deployer. Use of models at run-time and choreography of services for higher abstraction, adaptability and decentralization of the running system is proposed. Using this approach applications running in cloud environments may work based on their immediate needs.