Dynamic column in SELECT statement postgres

In order to write a dynamic query you would have to do something like:

EXECUTE 'SELECT '|| get_columns()|| ' FROM table_name' INTO results

Please read the documentation: http://developer.postgresql.org/pgdocs/postgres/plpgsql-statements.html

In such case I would use PL/pgSQL function using cursor.

You wont be able to use a function to generate a column list. And I really don't think this is the best way to approach the problem... That said, you can do it with 8.4 like so:

CREATE OR REPLACE FUNCTION dyn(p_name VARCHAR)
RETURNS SETOF RECORD AS
$$
  DECLARE
    p_sql  TEXT;
  BEGIN
   SELECT 'SELECT ' ||
     CASE p_name WHEN 'foo' THEN ' col1, col2, col3, col4 '
      WHEN 'bar' THEN 'col3, col4, col5, col6'
      WHEN 'baz' THEN 'col1, col3, col4, col6' END ||
   ' FROM mytest'
   INTO p_sql;
   RETURN QUERY EXECUTE p_sql;
  END
$$ LANGUAGE 'plpgsql';

Usage would be: SELECT * FROM dyn('foo') AS (one int, two int, three int, four int)

But honestly I'd suggest just making a view for each device.

Since you are using COPY FROM to a known large table, CREATE a FUNCTION which returns SETOF bigtable and SELECTs all the columns from the specific type, use NULL AS fieldname for the fields which are not required in that specific case, something like:

# \d SMALL
     Table "public.small"
 Column |  Type   | Modifiers 
--------+---------+-----------
 a      | integer | 
 b      | integer | 
 c      | integer | 
 d      | integer | 

# \d LARGE
     Table "public.large"
 Column |  Type   | Modifiers 
--------+---------+-----------
 a      | integer | 
 b      | integer | 
 c      | integer | 
 d      | integer | 

# CREATE OR REPLACE FUNCTION myData()
 RETURNS SETOF large LANGUAGE SQL AS $$
SELECT a, 
       CASE WHEN a = 1 
            THEN b 
       ELSE 
            NULL 
END as b, 
       CASE WHEN a = 2 
            THEN c 
       ELSE 
            NULL
END AS c, 
d
FROM small;
$$;

# SELECT * FROM mydata();
# COPY (SELECT * FROM myData()) TO STDOUT;

Obviously SQL might not be the best language to use, so PL/PgSQL or PL/Perl (or whatever) may be appropriate.

I think your biggest problem is that you need to return rows in a way that PostgreSQL can understand. This means basically, you can return a refcursor or you can return a consistent set of data types. I prefer the latter because it makes the system a bit more consistent from a programming perspective and there are some advanced directions you can take that but I can see the other way too.