What you are asking about is called a conditional trust where you tell a switch port to only trust specific devices. You can do this by device type:
you can use the command "mls qos trust device cisco-phone" at an interface level to instruct the switch to inspect CDP packets to ensure the device you are trusting is a phone. This is what the "auto qos voip cisco-phone" macro command is doing.
Another way to do a conditional trust is to apply an ingress policy on an interface. You create class-maps and policy-maps globally and then apply them to an interface. The "auto qos voip cisco-softphone" macro command actually creates the policy and applies it to the interface.
You can read up on what the auto qos commands are actually doing for you here:
http://www.cisco.com/en/US/docs/solutions/Enterprise/WAN_and_MAN/QoS_SRND/QoSDesign.html#wp999451
A later version is here. But the table in the above version is a little clearer.
http://www.cisco.com/en/US/docs/solutions/Enterprise/WAN_and_MAN/QoS_SRND_40/QoSCampus_40.html#wp1144082
As far as best practice, well that is a matter of opinion. The Cisco QoS SRND is a very good resource and worth the read. It will lay out several options for you. You need to understand your overall QoS design requirements and objectives before you go applying commands though. You have to think about voice, video, streaming, etc. But you also need to think about missiion critical data traffic that you or your company may care about.
Getting back to the network edge. Personally, I prefer to not trust on the edge. I like to use a class-map which leverages access lists to provide some granular control. More granular than what auto qos voip cisco-softphone gives your, to be sure. Anyway, just my opinion.
HTH.
Regards,
Bill
