The network models, OSI and TCP/IP, and their layers are conceptual abstractions that help show which level of "ordering" a communication component is operating at.
In practice, any application involved in network communication must be aware of and "operate" at the lower layers in order to work. By that I mean, the telnet application must know if it's using IPv4 or IPv6 and tcp/23 and the DNS application must know if it needs to use TCP or UDP. What they give to other layers and applications must be meaningful.
My analogy (for OSI):
Take a message from your mind (Application Layer), in your native language (Presentation), write it with an ink pen (Session), use multiple papers (Transport), put them all in an addressed envelope (Network) give it to the mail carrier (Data Link) and he carries to the mail address (Physical).
In this case, your mind (7) and pen (5) would create symbols understood by both your friend who reads your letter (7) and also the mail (3) carrier (2) who uses the address for delivery (1).
The layers are very distinct in text books but in practice can be very blurry.
The layers don't know about each other so much as the components (Apps, NICs) know what data is needed to get the job done. Your mind must be aware of your message, the pen, the paper, envelopes etc to get the message to your friend.
Does that help or have I made things more confusing? :)
