For communication overhead you have to consider which elements are present at each participant, and what is required at which time to perform the required operations. This is generally performed using a sequence diagram, where you note down which elements are communicated at specific steps of the protocol. For instance here is a simplified diagram for the Signal protocol.
The sequence diagram will show you the messages required and the elements together with their size indicate the communication overhead. Unless the protected messages are expanded the messages themself are of course not considered overhead, although they very much count towards the total bytes that need to be transmitted within a protocol. Very often the number of required messages is considered just as important, as that will also influence the latency for specific operations like connection setup.
Some elements like named parameter sets can be considered known within a protocol, or they take few bytes to communicate (just by indicating a version or an ID for the named parameter set). For instance, commonly DH and ECDH use a set of known parameters; if not those parameters need to be transmitted or at least reproduced at the receiver. Elements like private keys are of course never transmitted.
The word "ciphertext" is usually defined as the output of a cipher. Sometimes this includes an IV if that is prefixed. Sometimes it also includes an authentication tag if it is the output of an authenticated cipher. In the end it is up to the protocol to create a well defined definition. I'd however exclude things like a signature over the message; I'd use a term like "protected message" in case all the security overhead is included.